Methodology

How we score your brand.

The Robot Speed reputation report is a transparent quantitative audit. This page documents every choice, from the dimensions to the model ensemble and the confidence intervals. If something feels off, tell us — the methodology is a living thing.

7 cross-validated dimensions

Dimensions are cross-validated against RepTrak (7 dimensions), Fortune's World's Most Admired Companies (9 attributes), and Interbrand. Governance was deliberately dropped — LLMs hallucinate on private companies. For them, Leadership Visibility substitutes in, based on a measurable signal (CEO/founder public presence).

Salience

How spontaneously and frequently the brand is mentioned when AI discusses its category, unprompted.

0Unknown brand — AI requires prompting by name to discuss.
50Regional recognition, cited alongside 3-5 category alternatives.
100Category-defining brand cited first in any relevant prompt.

Innovation & Adaptability

Recent product launches, R&D, novel approaches, AI integration, ability to adapt to market shifts.

0No launches or meaningful updates in 2+ years.
50Follows category trends, incremental updates.
100Defines the frontier, launches copied by competitors.

Differentiation

How clearly and uniquely AI describes the brand's positioning vs. category peers. "What makes them X" vs "they are in X category".

0Described as generic category player, no distinct angle.
50Has a clear positioning but others claim similar.
100Owns a unique positioning AI attributes only to this brand.

Credibility

Perceived reliability, transparency, and trustworthiness based on AI-reported customer feedback, reviews, and public communications.

0Reports of scams, bad support, deceptive marketing.
50Mixed reviews, acceptable support, no major scandals.
100Industry reference for transparency and reliability.

Social Responsibility

Environmental commitments, social impact programs, ethical stances, DEI, sustainability signals.

0No ESG information or active controversies.
50Standard CSR commitments, nothing distinctive.
100Leading ESG commitments cited as category example.

Product & Service Quality

Craftsmanship, customer satisfaction with offerings, reliability of delivery, review sentiment.

0Consistent complaints about quality or broken features.
50Decent quality with known rough edges.
100Gold standard cited by competitors and reviews.

Market Position

Where AI ranks the brand against competitors — top choice, alternative, niche, or also-ran.

0Not mentioned when AI ranks top 10 in category.
50Cited as alternative or mid-market option.
100Consistently ranked #1 or #2 by AI across prompts.

Two-judge model ensemble

Each dimension is scored 6 times per report — 2 independent judge models (2 distinct model families) × 3 runs with staggered temperatures (0.3 / 0.6 / 0.9). We return mean ± one standard deviation. An SD above 15 points triggers the "low consensus" flag — a signal that judges disagree. We don't smooth uncertainty to look confident.

Evidence: CMU paper (Dec 2025) on LLM-as-judge rating indeterminacy; Wang et al. (2024) on position-bias mitigation via prompt-order averaging; Multi-Agent Verification (Feb 2025) on ensemble superiority over model scaling.

Calibrated anchors

The scoring prompt includes anchor examples at 0, 50, and 100 for each dimension to prevent the polite clustering around 60-80 that LLMs tend to produce. We want honest variance — an average brand comes out at 45, not 70.

Data sourcing

We query ChatGPT, Perplexity, and Google AI Overviews via an established data-scraping proxy (the same kind of infrastructure used by Semrush, Ahrefs, and others). Each provider gets the same structured prompt asking for a 7-dimension analysis with verbatim quotes. Responses are time-boxed to 50s per provider — if one times out, we continue with the rest and show a "preliminary report" banner.

Access and transparency

The tool is free for you, with a 10-reports-per-hour-per-IP rate limit. Beyond that, you hit the fair-use ceiling and we redirect you to the paid tier (weekly tracking, drop alerts, 3/6/12-month history).

Honest limitations

(1) LLMs hallucinate — we don't measure this yet in production (Factual Accuracy Layer shipping soon). (2) Very recent or narrowly regional brands aren't always known — we return raw LLM responses even when we can't score. (3) The 6-run average reduces noise but not systemic bias — a model that thinks a brand is bad on a dimension usually keeps thinking so. Cross-vendor (2 distinct model families) mitigates.

Benchmarked against market tools

Most competitors (LLMPulse, Otterly, Peec AI) don't publish their formulas. We do. If you find a bias we missed, tell [email protected].